Potential problems of applying the model-oriented paradigm in mobile network design

DOI: 10.31673/2412-9070.2025.017038

Abstract

In this paper, we address the limitations of the traditional model-based paradigm in mobile wireless communication networks, such as the complexity of defining accurate models, obtaining system parameters, high computational demands, and the inability to create lossless block decompositions. These challenges hinder the effective use of model-based approaches in dynamic, evolving mobile networks.
We analyze the data-driven paradigm, supported by advanced machine learning techniques, as a solution. Unlike model-based approaches, which depend on predefined network models and parameters, the data-driven paradigm builds networks directly on data generated by the network itself. This approach efficiently handles real-time, dynamic data, bypassing the need for static models. We investigate one of a typical use case where this paradigm is applied to implement proactive load balancing. A key feature of this approach is online learning, which enables networks to predict traffic spikes before they occur and adjust network parameters accordingly. This proactive strategy minimizes congestion and improves efficiency by anticipating traffic surges, such as those caused by big groups of users which are moving between base stations. The paper describes how online learning methods are used to predict and avoid packet overloads due to rapid traffic changes, especially in mobile networks with varying data rates. We also investigated load balancing using online learning, where access points independently predict traffic based on neighboring cells' data, and a central load balancer adjusts cell configurations to reduce congestion. Proactive adjustments are made before traffic spikes, ensuring minimized disruptions.
In conclusion, the data-driven paradigm, combined with machine learning, offers significant advantages over traditional approaches, particularly in scalability, flexibility, and real-time adaptability. As mobile networks evolve, further research into these methods will be crucial for more efficient, intelligent communication systems.

Keywords: 5G, model-based approach, data, mobile wireless networks, machine learning, traffic prediction, load balancing, computational complexity.